Home

 › 

Uncategorized

 › 

‘They’re Not Trying to Create God in a Box.’ A Shanghai-Based Advisor Says America’s GPU Lead Is Real, and Still Woefully Incomplete.

A panoramic view from a high-rise office shows the illuminated Shanghai skyline at dusk, with the Oriental Pearl TV Tower and other skyscrapers alongside the Huangpu River. In the office, a man in a suit sits at a desk, looking at a large wall screen displaying a 'Global GPU Market Share' pie chart with a 70-75% segment and an image of robotic arms in a factory. A tablet in the foreground displays a quote: 'They're not trying to create god in a box.' and an illustration of a microchip.

‘They’re Not Trying to Create God in a Box.’ A Shanghai-Based Advisor Says America’s GPU Lead Is Real, and Still Woefully Incomplete.

Quick Read

  • America leads China in GPUs by a wide margin, yet one Shanghai-based advisor argues that the scoreboard is measuring the wrong race entirely.
  • Chinese AI firms are giving their models away for free, and doing so might actually be a smarter business plan than selling the most powerful one.
  • The biggest vulnerability in China's AI strategy hinges on a single assumption, and that assumption could unravel everything if frontier models keep improving.
  • Many financial professionals are salespeople paid on what they push, not whether you end up wealthier. A fiduciary is the opposite. The SEC legally requires them to put your interests first. Advisor.com's free matching tool pairs you with vetted fiduciaries from major national firms, all in under three minutes. See who you match with today.

Most assessments of the US-China AI contest start with hardware, where America leads. Peter Alexander, founder and managing director of Shanghai-based Z-Ben Advisors accepts that lead. It argues the standard measure captures only part of the competition. The timing matters: the two countries’ AI rivalry has dominated headlines this week, shaping how investors frame the contest.

He Grants America’s Lead Before Challenging It

Alexander opened by agreeing with the premise. “What you have is a situation where right now what you just stated is absolutely correct. It is just a position that is woefully incomplete.”

He notes the US holds roughly 70% to 75% of global GPUs and is ahead on frontier models. His challenge is whether that scoreboard tracks the contest China has chosen to play.

He condensed the whole argument into one image:

“The Chinese open models, they’re not trying to create god in a box. The competition is not exactly the same.”

Two Races With Different Finish Lines

Alexander described the actual objective: “What the Chinese open models are attempting to do is apply for application level usage of the AI through manufacturing and elsewhere, and they have been doing an exceptional job on that front.”

He then stated the strategy directly. The strategy was “The Chinese know that their AI is not matching frontier and is behind, but good enough is good enough for what it is they want to produce.”

Two competitors can share a race and run toward different finish lines, with one chasing maximum general capability, measured by frontier benchmarks. The other chases deployment into industrial processes where a less capable model serves. On the first measure, the US is winning. On the second, the outcome remains wide open.

Why Giving Models Away Can Be a Business Plan

Chinese AI companies use open source and open weight models to drive global adoption. An open-weight model releases trained parameters publicly, so anyone can download, run and modify it. Chinese firms prioritise market share over profit margins.

The playbook mirrors software platforms: give the product away, build the installed base, monetize later, and applied to AI, it competes on a separate axis from selling paid access to the most capable model. One profits on capability. The other earns on ubiquity, and ubiquity inside factories can be sticky once workflows are built around it.

His Read on US Policy Toward China

Alexander views US policy toward China as stressing containment and treating China as an enemy instead of exploring cooperation. That is his opinion, offered as part of his broader argument.

His vantage point matters: Alexander runs a Shanghai-based advisory firm, so his professional lens is the Chinese market.

The Condition His Thesis Depends On

His argument assumes “good enough” describes both the present and future, and the thesis requires it to keep describing the future.

His case holds if the frontier premium stays confined to the frontier. If more capable general models produce significantly better results inside manufacturing, an application-layer advantage built on weaker models would erode. Alexander’s position requires that the capability gap at the top does not extend to into the use cases he describes.

This represents the testable part of his thesis. His reframing of the scoreboard stands on its own. What it cannot establish by itself is that good enough stays good enough.

Two Scoreboards Worth Tracking

The durable insight is the reframing. Measuring AI competition by GPU counts and frontier benchmarks answers one question: who can build the most capable systems. Measuring it by how much industrial production actually runs on AI answers another: who is putting the technology to work at scale.

Both metrics are real, and they can point in opposite directions for a long time. For investors following the sector, the signal to monitor is whether gains at the frontier show up on the factory floor. That evidence would reveal which scoreboard carries more weight.

To top